Paper accepted at ECCV 2016
My latest paper entitled Region-based semantic segmentation with end-to-end training has been accepted the the European Conference on Computer Vision (ECCV) 2016. Semantic segmentation is the task of labeling each pixel in an image. Our work combines the advantages of the two main paradigms in semantic segmentation: Fully Convolutional Neural Networks can be trained end-to-end and region-based methods allow especially crisp object boundaries. Our method achieves state-of-the-art results in class-average pixel accuracy on both the SIFT Flow and PASCAL Context datasets.